Private Writing AssistantDo AI writing tools train on what you type?
It depends on the tool, the plan you're on and your settings. Some services use your text to improve their models by default, some only with your permission, and some never. The privacy policy and the tool's data settings tell you which, and a tool that runs on your own device can't train on text it never receives.
Why the answer isn't simply yes or no
“Training” covers several different things. A company might use your text to train the next version of its model, to fine-tune a smaller model for one feature, to have staff review samples for quality, or only to run your request and then delete it. Policies often treat these separately.
The answer also changes with your account. Many services treat free personal accounts differently from paid business plans, and some let you opt out in settings. The same tool can train on one person's text and not on another's.
Where to look in the privacy policy
Search the policy for these words, and read the paragraphs around them:
- “improve our services” or “develop new features”: the usual wording when your content is used for training or review.
- “content”, “inputs” or “prompts”: how the policy names the text you type, as opposed to account or usage data.
- “human review” or “reviewers”: whether people may read samples of what you write.
- “retain” or “retention”: how long your text is kept, which matters even if it is never used for training.
- “opt out”: whether you can say no, and whether no is the default.
If the policy talks only about “data” without saying whether that includes what you write, assume it might. A policy that excludes your text usually says so plainly.
Where to look in the settings
Policies describe what a company may do; settings show what applies to you. Look in the account or privacy settings for a switch about model improvement, training or sharing data. Check it on every account you use, including the one on your work laptop, because the defaults may differ.
Note the date when you check. Policies and defaults change, and a service that didn't train on your text last year may ask you to opt out this year.
What an opt-out doesn't cover
Turning off training doesn't mean your text stays with you. A cloud tool still has to receive your text to work on it, so it still crosses the internet, sits on the company's servers for some time, and falls under their security, their staff and the laws of the countries where they operate.
For most messages that's an acceptable trade. For a client's contract, a patient's details or an unannounced plan, it's worth asking whether the text needs to leave your computer at all. Our post Is it safe to paste confidential text into AI? goes through that decision.
The question that settles it
Instead of asking whether a tool trains on your text, ask whether it receives your text at all. A tool that processes text on your own device has nothing to train on, keep or review, because it never gets a copy. That is a property of how it's built, not a promise you have to trust or a setting you have to find.
To check that claim, look for three things: the tool keeps working offline, it needs a model download or uses one your browser provides, and its policy says where your text is processed. Our post What does “on-device AI” actually mean? explains the difference in more detail.
How Private Writing Assistant handles it
Private Writing Assistant checks and rewrites text with Chrome Built-in AI on your computer. What you write on websites is never sent to us or to an AI service, so it can't be used to train any model. Two things can leave your device, and neither contains your text: usage stats, which you can turn off in settings, and feedback you choose to send. The privacy policy lists every detail.
Check spelling and grammar and rewrite text without sending it anywhere: add Private Writing Assistant to Chrome. It's free and needs no account.
